Details
-
Bug
-
Status: Closed (View Workflow)
-
Critical
-
Resolution: Fixed
-
10.11, 11.4, 11.8, 12.0(EOL)
-
None
Description
Test failure:
main.mdl_sync w25 [ fail ]
|
Test ended at 2025-04-21 01:50:53
|
|
|
CURRENT_TEST: main.mdl_sync
|
--- /home/buildbot/amd64-fedora-40-valgrind/build/mysql-test/main/mdl_sync.result 2025-04-19 16:41:39.000000000 +0000
|
+++ /home/buildbot/amd64-fedora-40-valgrind/build/mysql-test/main/mdl_sync.reject 2025-04-21 01:50:52.064219508 +0000
|
@@ -2655,6 +2655,13 @@
|
SET debug_sync='now WAIT_FOR parked_flush';
|
SET debug_sync='now SIGNAL go_truncate';
|
# Ensure that truncate waits for a exclusive lock
|
+Timeout in wait_condition.inc for SELECT COUNT(*)=1 FROM information_schema.processlist
|
+WHERE state='Waiting for table metadata lock' AND info='TRUNCATE TABLE t1'
|
+Id User Host db Command Time State Info Progress
|
+4 root localhost test Query 0 starting show full processlist 0.000
|
+34 root localhost test Sleep 33 NULL 0.000
|
+35 root localhost test Sleep 33 NULL 0.000
|
+36 root localhost test Query 450 Waiting for table metadata lock FLUSH TABLES t1 0.000
|
SET debug_sync= 'now SIGNAL go_show';
|
connection con1;
|
# Reaping...
|
@@ -2663,10 +2670,14 @@
|
# Reaping...
|
Field Type Null Key Default Extra
|
a int(11) YES NULL
|
+Warnings:
|
+Warning 1639 debug sync point wait timed out
|
connection default;
|
SET debug_sync= 'now SIGNAL go_flush';
|
connection con3;
|
# Reaping...
|
+Warnings:
|
+Warning 1639 debug sync point wait timed out
|
disconnect con1;
|
disconnect con2;
|
disconnect con3;
|
|
|
Result length mismatch
|
|
|
- saving '/home/buildbot/amd64-fedora-40-valgrind/build/mysql-test/var/25/log/main.mdl_sync/' to '/home/buildbot/amd64-fedora-40-valgrind/build/mysql-test/var/log/main.mdl_sync/'
|
The problem here appears to be:
1. TRUNCATE (after "now SIGNAL go_truncate") starts waiting for MDL_EXCLUSIVE lock on t1
2. InnoDB purge thread chimes in, attempts taking MDL_SHARED lock on t1, fails, retries in a loop
Given valgrind scheduling specifics, InnoDB purge thread occupies the whole CPU and never yields CPU to user connections, causing sync point timeout.
Fix for this issue was proposed a while ago: https://github.com/MariaDB/server/commit/0c6c580137146492e234570df30d302cafd94131
Test for Bug#42643 consistently failing without the fix (mtr --repeat=100 --valgrind --parallel=20). No failures were observed with the fix.
Attachments
Issue Links
- is part of
-
MDEV-36647 No red leaves in the forest
-
- Open
-